Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / mm / backing-dev.c
1
2 #include <linux/wait.h>
3 #include <linux/backing-dev.h>
4 #include <linux/kthread.h>
5 #include <linux/freezer.h>
6 #include <linux/fs.h>
7 #include <linux/pagemap.h>
8 #include <linux/mm.h>
9 #include <linux/sched.h>
10 #include <linux/module.h>
11 #include <linux/writeback.h>
12 #include <linux/device.h>
13 #include <trace/events/writeback.h>
14
15 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
16
17 struct backing_dev_info default_backing_dev_info = {
18 .name = "default",
19 .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
20 .state = 0,
21 .capabilities = BDI_CAP_MAP_COPY,
22 };
23 EXPORT_SYMBOL_GPL(default_backing_dev_info);
24
25 struct backing_dev_info noop_backing_dev_info = {
26 .name = "noop",
27 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
28 };
29 EXPORT_SYMBOL_GPL(noop_backing_dev_info);
30
31 static struct class *bdi_class;
32
33 /*
34 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
35 * locking.
36 */
37 DEFINE_SPINLOCK(bdi_lock);
38 LIST_HEAD(bdi_list);
39
40 /* bdi_wq serves all asynchronous writeback tasks */
41 struct workqueue_struct *bdi_wq;
42
43 void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2)
44 {
45 if (wb1 < wb2) {
46 spin_lock(&wb1->list_lock);
47 spin_lock_nested(&wb2->list_lock, 1);
48 } else {
49 spin_lock(&wb2->list_lock);
50 spin_lock_nested(&wb1->list_lock, 1);
51 }
52 }
53
54 #ifdef CONFIG_DEBUG_FS
55 #include <linux/debugfs.h>
56 #include <linux/seq_file.h>
57
58 static struct dentry *bdi_debug_root;
59
60 static void bdi_debug_init(void)
61 {
62 bdi_debug_root = debugfs_create_dir("bdi", NULL);
63 }
64
65 static int bdi_debug_stats_show(struct seq_file *m, void *v)
66 {
67 struct backing_dev_info *bdi = m->private;
68 struct bdi_writeback *wb = &bdi->wb;
69 unsigned long background_thresh;
70 unsigned long dirty_thresh;
71 unsigned long bdi_thresh;
72 unsigned long nr_dirty, nr_io, nr_more_io;
73 struct inode *inode;
74
75 nr_dirty = nr_io = nr_more_io = 0;
76 spin_lock(&wb->list_lock);
77 list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
78 nr_dirty++;
79 list_for_each_entry(inode, &wb->b_io, i_wb_list)
80 nr_io++;
81 list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
82 nr_more_io++;
83 spin_unlock(&wb->list_lock);
84
85 global_dirty_limits(&background_thresh, &dirty_thresh);
86 bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
87
88 #define K(x) ((x) << (PAGE_SHIFT - 10))
89 seq_printf(m,
90 "BdiWriteback: %10lu kB\n"
91 "BdiReclaimable: %10lu kB\n"
92 "BdiDirtyThresh: %10lu kB\n"
93 "DirtyThresh: %10lu kB\n"
94 "BackgroundThresh: %10lu kB\n"
95 "BdiDirtied: %10lu kB\n"
96 "BdiWritten: %10lu kB\n"
97 "BdiWriteBandwidth: %10lu kBps\n"
98 "b_dirty: %10lu\n"
99 "b_io: %10lu\n"
100 "b_more_io: %10lu\n"
101 "bdi_list: %10u\n"
102 "state: %10lx\n",
103 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
104 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
105 K(bdi_thresh),
106 K(dirty_thresh),
107 K(background_thresh),
108 (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)),
109 (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
110 (unsigned long) K(bdi->write_bandwidth),
111 nr_dirty,
112 nr_io,
113 nr_more_io,
114 !list_empty(&bdi->bdi_list), bdi->state);
115 #undef K
116
117 return 0;
118 }
119
120 static int bdi_debug_stats_open(struct inode *inode, struct file *file)
121 {
122 return single_open(file, bdi_debug_stats_show, inode->i_private);
123 }
124
125 static const struct file_operations bdi_debug_stats_fops = {
126 .open = bdi_debug_stats_open,
127 .read = seq_read,
128 .llseek = seq_lseek,
129 .release = single_release,
130 };
131
132 static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
133 {
134 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
135 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
136 bdi, &bdi_debug_stats_fops);
137 }
138
139 static void bdi_debug_unregister(struct backing_dev_info *bdi)
140 {
141 debugfs_remove(bdi->debug_stats);
142 debugfs_remove(bdi->debug_dir);
143 }
144 #else
145 static inline void bdi_debug_init(void)
146 {
147 }
148 static inline void bdi_debug_register(struct backing_dev_info *bdi,
149 const char *name)
150 {
151 }
152 static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
153 {
154 }
155 #endif
156
157 static ssize_t read_ahead_kb_store(struct device *dev,
158 struct device_attribute *attr,
159 const char *buf, size_t count)
160 {
161 struct backing_dev_info *bdi = dev_get_drvdata(dev);
162 unsigned long read_ahead_kb;
163 ssize_t ret;
164
165 ret = kstrtoul(buf, 10, &read_ahead_kb);
166 if (ret < 0)
167 return ret;
168
169 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
170
171 return count;
172 }
173
174 #define K(pages) ((pages) << (PAGE_SHIFT - 10))
175
176 #define BDI_SHOW(name, expr) \
177 static ssize_t name##_show(struct device *dev, \
178 struct device_attribute *attr, char *page) \
179 { \
180 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
181 \
182 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
183 }
184
185 BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
186
187 static ssize_t min_ratio_store(struct device *dev,
188 struct device_attribute *attr, const char *buf, size_t count)
189 {
190 struct backing_dev_info *bdi = dev_get_drvdata(dev);
191 unsigned int ratio;
192 ssize_t ret;
193
194 ret = kstrtouint(buf, 10, &ratio);
195 if (ret < 0)
196 return ret;
197
198 ret = bdi_set_min_ratio(bdi, ratio);
199 if (!ret)
200 ret = count;
201
202 return ret;
203 }
204 BDI_SHOW(min_ratio, bdi->min_ratio)
205
206 static ssize_t max_ratio_store(struct device *dev,
207 struct device_attribute *attr, const char *buf, size_t count)
208 {
209 struct backing_dev_info *bdi = dev_get_drvdata(dev);
210 unsigned int ratio;
211 ssize_t ret;
212
213 ret = kstrtouint(buf, 10, &ratio);
214 if (ret < 0)
215 return ret;
216
217 ret = bdi_set_max_ratio(bdi, ratio);
218 if (!ret)
219 ret = count;
220
221 return ret;
222 }
223 BDI_SHOW(max_ratio, bdi->max_ratio)
224
225 static ssize_t stable_pages_required_show(struct device *dev,
226 struct device_attribute *attr,
227 char *page)
228 {
229 struct backing_dev_info *bdi = dev_get_drvdata(dev);
230
231 return snprintf(page, PAGE_SIZE-1, "%d\n",
232 bdi_cap_stable_pages_required(bdi) ? 1 : 0);
233 }
234
235 #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
236
237 static struct device_attribute bdi_dev_attrs[] = {
238 __ATTR_RW(read_ahead_kb),
239 __ATTR_RW(min_ratio),
240 __ATTR_RW(max_ratio),
241 __ATTR_RO(stable_pages_required),
242 __ATTR_NULL,
243 };
244
245 static __init int bdi_class_init(void)
246 {
247 bdi_class = class_create(THIS_MODULE, "bdi");
248 if (IS_ERR(bdi_class))
249 return PTR_ERR(bdi_class);
250
251 bdi_class->dev_attrs = bdi_dev_attrs;
252 bdi_debug_init();
253 return 0;
254 }
255 postcore_initcall(bdi_class_init);
256
257 static int __init default_bdi_init(void)
258 {
259 int err;
260
261 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
262 WQ_UNBOUND | WQ_SYSFS, 0);
263 if (!bdi_wq)
264 return -ENOMEM;
265
266 err = bdi_init(&default_backing_dev_info);
267 if (!err)
268 bdi_register(&default_backing_dev_info, NULL, "default");
269 err = bdi_init(&noop_backing_dev_info);
270
271 return err;
272 }
273 subsys_initcall(default_bdi_init);
274
275 int bdi_has_dirty_io(struct backing_dev_info *bdi)
276 {
277 return wb_has_dirty_io(&bdi->wb);
278 }
279
280 /*
281 * This function is used when the first inode for this bdi is marked dirty. It
282 * wakes-up the corresponding bdi thread which should then take care of the
283 * periodic background write-out of dirty inodes. Since the write-out would
284 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
285 * set up a timer which wakes the bdi thread up later.
286 *
287 * Note, we wouldn't bother setting up the timer, but this function is on the
288 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
289 * by delaying the wake-up.
290 */
291 void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
292 {
293 unsigned long timeout;
294
295 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
296 mod_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
297 }
298
299 /*
300 * Remove bdi from bdi_list, and ensure that it is no longer visible
301 */
302 static void bdi_remove_from_list(struct backing_dev_info *bdi)
303 {
304 spin_lock_bh(&bdi_lock);
305 list_del_rcu(&bdi->bdi_list);
306 spin_unlock_bh(&bdi_lock);
307
308 synchronize_rcu_expedited();
309
310 /* bdi_list is now unused, clear it to mark @bdi dying */
311 INIT_LIST_HEAD(&bdi->bdi_list);
312 }
313
314 int bdi_register(struct backing_dev_info *bdi, struct device *parent,
315 const char *fmt, ...)
316 {
317 va_list args;
318 struct device *dev;
319
320 if (bdi->dev) /* The driver needs to use separate queues per device */
321 return 0;
322
323 va_start(args, fmt);
324 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
325 va_end(args);
326 if (IS_ERR(dev))
327 return PTR_ERR(dev);
328
329 bdi->dev = dev;
330
331 bdi_debug_register(bdi, dev_name(dev));
332 set_bit(BDI_registered, &bdi->state);
333
334 spin_lock_bh(&bdi_lock);
335 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
336 spin_unlock_bh(&bdi_lock);
337
338 trace_writeback_bdi_register(bdi);
339 return 0;
340 }
341 EXPORT_SYMBOL(bdi_register);
342
343 int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
344 {
345 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
346 }
347 EXPORT_SYMBOL(bdi_register_dev);
348
349 /*
350 * Remove bdi from the global list and shutdown any threads we have running
351 */
352 static void bdi_wb_shutdown(struct backing_dev_info *bdi)
353 {
354 if (!bdi_cap_writeback_dirty(bdi))
355 return;
356
357 /*
358 * Make sure nobody finds us on the bdi_list anymore
359 */
360 bdi_remove_from_list(bdi);
361
362 /*
363 * Drain work list and shutdown the delayed_work. At this point,
364 * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
365 * is dying and its work_list needs to be drained no matter what.
366 */
367 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
368 flush_delayed_work(&bdi->wb.dwork);
369 WARN_ON(!list_empty(&bdi->work_list));
370
371 /*
372 * This shouldn't be necessary unless @bdi for some reason has
373 * unflushed dirty IO after work_list is drained. Do it anyway
374 * just in case.
375 */
376 cancel_delayed_work_sync(&bdi->wb.dwork);
377 }
378
379 /*
380 * This bdi is going away now, make sure that no super_blocks point to it
381 */
382 static void bdi_prune_sb(struct backing_dev_info *bdi)
383 {
384 struct super_block *sb;
385
386 spin_lock(&sb_lock);
387 list_for_each_entry(sb, &super_blocks, s_list) {
388 if (sb->s_bdi == bdi)
389 sb->s_bdi = &default_backing_dev_info;
390 }
391 spin_unlock(&sb_lock);
392 }
393
394 void bdi_unregister(struct backing_dev_info *bdi)
395 {
396 struct device *dev = bdi->dev;
397
398 if (dev) {
399 bdi_set_min_ratio(bdi, 0);
400 trace_writeback_bdi_unregister(bdi);
401 bdi_prune_sb(bdi);
402
403 bdi_wb_shutdown(bdi);
404 bdi_debug_unregister(bdi);
405
406 spin_lock_bh(&bdi->wb_lock);
407 bdi->dev = NULL;
408 spin_unlock_bh(&bdi->wb_lock);
409
410 device_unregister(dev);
411 }
412 }
413 EXPORT_SYMBOL(bdi_unregister);
414
415 static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
416 {
417 memset(wb, 0, sizeof(*wb));
418
419 wb->bdi = bdi;
420 wb->last_old_flush = jiffies;
421 INIT_LIST_HEAD(&wb->b_dirty);
422 INIT_LIST_HEAD(&wb->b_io);
423 INIT_LIST_HEAD(&wb->b_more_io);
424 spin_lock_init(&wb->list_lock);
425 INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
426 }
427
428 /*
429 * Initial write bandwidth: 100 MB/s
430 */
431 #define INIT_BW (100 << (20 - PAGE_SHIFT))
432
433 int bdi_init(struct backing_dev_info *bdi)
434 {
435 int i, err;
436
437 bdi->dev = NULL;
438
439 bdi->min_ratio = 0;
440 bdi->max_ratio = 100;
441 bdi->max_prop_frac = FPROP_FRAC_BASE;
442 spin_lock_init(&bdi->wb_lock);
443 INIT_LIST_HEAD(&bdi->bdi_list);
444 INIT_LIST_HEAD(&bdi->work_list);
445
446 bdi_wb_init(&bdi->wb, bdi);
447
448 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
449 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
450 if (err)
451 goto err;
452 }
453
454 bdi->dirty_exceeded = 0;
455
456 bdi->bw_time_stamp = jiffies;
457 bdi->written_stamp = 0;
458
459 bdi->balanced_dirty_ratelimit = INIT_BW;
460 bdi->dirty_ratelimit = INIT_BW;
461 bdi->write_bandwidth = INIT_BW;
462 bdi->avg_write_bandwidth = INIT_BW;
463
464 err = fprop_local_init_percpu(&bdi->completions);
465
466 if (err) {
467 err:
468 while (i--)
469 percpu_counter_destroy(&bdi->bdi_stat[i]);
470 }
471
472 return err;
473 }
474 EXPORT_SYMBOL(bdi_init);
475
476 void bdi_destroy(struct backing_dev_info *bdi)
477 {
478 int i;
479
480 /*
481 * Splice our entries to the default_backing_dev_info, if this
482 * bdi disappears
483 */
484 if (bdi_has_dirty_io(bdi)) {
485 struct bdi_writeback *dst = &default_backing_dev_info.wb;
486
487 bdi_lock_two(&bdi->wb, dst);
488 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
489 list_splice(&bdi->wb.b_io, &dst->b_io);
490 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
491 spin_unlock(&bdi->wb.list_lock);
492 spin_unlock(&dst->list_lock);
493 }
494
495 bdi_unregister(bdi);
496
497 /*
498 * If bdi_unregister() had already been called earlier, the dwork
499 * could still be pending because bdi_prune_sb() can race with the
500 * bdi_wakeup_thread_delayed() calls from __mark_inode_dirty().
501 */
502 cancel_delayed_work_sync(&bdi->wb.dwork);
503
504 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
505 percpu_counter_destroy(&bdi->bdi_stat[i]);
506
507 fprop_local_destroy_percpu(&bdi->completions);
508 }
509 EXPORT_SYMBOL(bdi_destroy);
510
511 /*
512 * For use from filesystems to quickly init and register a bdi associated
513 * with dirty writeback
514 */
515 int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
516 unsigned int cap)
517 {
518 char tmp[32];
519 int err;
520
521 bdi->name = name;
522 bdi->capabilities = cap;
523 err = bdi_init(bdi);
524 if (err)
525 return err;
526
527 sprintf(tmp, "%.28s%s", name, "-%d");
528 err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
529 if (err) {
530 bdi_destroy(bdi);
531 return err;
532 }
533
534 return 0;
535 }
536 EXPORT_SYMBOL(bdi_setup_and_register);
537
538 static wait_queue_head_t congestion_wqh[2] = {
539 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
540 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
541 };
542 static atomic_t nr_bdi_congested[2];
543
544 void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
545 {
546 enum bdi_state bit;
547 wait_queue_head_t *wqh = &congestion_wqh[sync];
548
549 bit = sync ? BDI_sync_congested : BDI_async_congested;
550 if (test_and_clear_bit(bit, &bdi->state))
551 atomic_dec(&nr_bdi_congested[sync]);
552 smp_mb__after_clear_bit();
553 if (waitqueue_active(wqh))
554 wake_up(wqh);
555 }
556 EXPORT_SYMBOL(clear_bdi_congested);
557
558 void set_bdi_congested(struct backing_dev_info *bdi, int sync)
559 {
560 enum bdi_state bit;
561
562 bit = sync ? BDI_sync_congested : BDI_async_congested;
563 if (!test_and_set_bit(bit, &bdi->state))
564 atomic_inc(&nr_bdi_congested[sync]);
565 }
566 EXPORT_SYMBOL(set_bdi_congested);
567
568 /**
569 * congestion_wait - wait for a backing_dev to become uncongested
570 * @sync: SYNC or ASYNC IO
571 * @timeout: timeout in jiffies
572 *
573 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
574 * write congestion. If no backing_devs are congested then just wait for the
575 * next write to be completed.
576 */
577 long congestion_wait(int sync, long timeout)
578 {
579 long ret;
580 unsigned long start = jiffies;
581 DEFINE_WAIT(wait);
582 wait_queue_head_t *wqh = &congestion_wqh[sync];
583
584 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
585 ret = io_schedule_timeout(timeout);
586 finish_wait(wqh, &wait);
587
588 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
589 jiffies_to_usecs(jiffies - start));
590
591 return ret;
592 }
593 EXPORT_SYMBOL(congestion_wait);
594
595 /**
596 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
597 * @zone: A zone to check if it is heavily congested
598 * @sync: SYNC or ASYNC IO
599 * @timeout: timeout in jiffies
600 *
601 * In the event of a congested backing_dev (any backing_dev) and the given
602 * @zone has experienced recent congestion, this waits for up to @timeout
603 * jiffies for either a BDI to exit congestion of the given @sync queue
604 * or a write to complete.
605 *
606 * In the absence of zone congestion, cond_resched() is called to yield
607 * the processor if necessary but otherwise does not sleep.
608 *
609 * The return value is 0 if the sleep is for the full timeout. Otherwise,
610 * it is the number of jiffies that were still remaining when the function
611 * returned. return_value == timeout implies the function did not sleep.
612 */
613 long wait_iff_congested(struct zone *zone, int sync, long timeout)
614 {
615 long ret;
616 unsigned long start = jiffies;
617 DEFINE_WAIT(wait);
618 wait_queue_head_t *wqh = &congestion_wqh[sync];
619
620 /*
621 * If there is no congestion, or heavy congestion is not being
622 * encountered in the current zone, yield if necessary instead
623 * of sleeping on the congestion queue
624 */
625 if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
626 !zone_is_reclaim_congested(zone)) {
627 cond_resched();
628
629 /* In case we scheduled, work out time remaining */
630 ret = timeout - (jiffies - start);
631 if (ret < 0)
632 ret = 0;
633
634 goto out;
635 }
636
637 /* Sleep until uncongested or a write happens */
638 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
639 ret = io_schedule_timeout(timeout);
640 finish_wait(wqh, &wait);
641
642 out:
643 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
644 jiffies_to_usecs(jiffies - start));
645
646 return ret;
647 }
648 EXPORT_SYMBOL(wait_iff_congested);
649
650 int pdflush_proc_obsolete(struct ctl_table *table, int write,
651 void __user *buffer, size_t *lenp, loff_t *ppos)
652 {
653 char kbuf[] = "0\n";
654
655 if (*ppos) {
656 *lenp = 0;
657 return 0;
658 }
659
660 if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
661 return -EFAULT;
662 printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
663 table->procname);
664
665 *lenp = 2;
666 *ppos += *lenp;
667 return 2;
668 }